1

Our AWS configuration has an nginx server pointing all PHP requests to an ELB with multiple PHP servers in it. When the ELB itself scales, we lose connectivity within the next few hours as those old IPs are recycled. A restart of the nginx servers re-caches the IPs of the ELB, and everything comes back up. Pretty standard stuff.

The solution I've found everywhere is that I need to include a resolver to allow re-resolution while the server is up and running. The issue I'm running into is two-fold:

  1. My current configuration isn't working.
  2. I'm not sure how to properly test it, without waiting for it to fail.

Here's my current configuration:

http {
    resolver 10.0.0.2 valid=10s;
    ...

    server {
        ...

        location ~ \.php$ {
            include /etc/nginx/fastcgi_params;
            fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
            set $route "php:9000";
            fastcgi_pass $route;
        }
    }
}

An article I found mentioned using tcpdump to review DNS requests, but I don't see anything at all, and definitely not every 10 seconds as written. Could it have to do with using fastcgi_pass instead of proxy_pass? Am I missing something else?

ahamilton9
  • 13
  • 4
  • Can you expand your question to describe the problem and symptoms more fully? I'm reasonably good with Nginx, not quite as good with networks, and I don't fully understand your question. – Tim Nov 21 '16 at 18:06
  • I'll add some detail above, but basically, I'm passing PHP files to an internal load balancer via fastcgi_pass. About once a week I need to manually restart the servers due to some issue, which I'm under the assumption has to do with the DNS re-resolution. – ahamilton9 Nov 21 '16 at 18:24
  • I'd look more closely at that assumption, you can use something like Smokeping to monitor. You could also just use IP addresses if the problem is resolving internal server names. – Tim Nov 21 '16 at 18:26
  • I can't just use IP addresses, as the load balancer's IPs may change when the systems in it scale up or down. I will continue to monitor the IPs, but I've been looking into this for a while now, and have yet to find any other possibility. – ahamilton9 Nov 21 '16 at 18:31

0 Answers0